Professional Documents
Culture Documents
ABSTRACT
We present a new variant of vehicle routing problem with a min-max objective
function. The problem has different types of service demands satisfied by a heterogeneous
fleet of vehicles. Unlimited service capacitated vehicles serve the demand points with
multiple time windows and requirement of operation synchronization when demand is split
between vehicles. A mixed integer linear programming based heuristic solution approach is
proposed and a numerical study is carried out to assess the performance of the proposed
method.
ÖZ
Bu çalışmada min-max amaç fonksiyonlu yeni bir çeşit araç rotalama problemi
sunulmaktadır. Problemde farklı tipte araçlardan oluşan bir filo ile farklı türde hizmet
talepleri karşılanmaktadır. Hizmet kapasitesi sınırsız olan araçlar, birden fazla zaman
pencereli ve talebin araçlar arasında bölünerek karşılanması halinde operasyon
senkronizasyonu gerektiren talep noktalarına hizmet sağlamaktadır. Problemin çözümü
için tamsayılı doğrusal programlama tabanlı bir sezgisel algoritma yaklaşımı önerilmiş ve
önerilen metodun persformansını değerlendirmek için sayısal bir çalışma yapılmıştır.
Keywords: Vehicle Routing Problem; Heuristic Method; Synchronization.
Anahtar Kelimeler: Araç Rotalama Problemi; Sezgisel Yöntemler; Senkronizasyon.
1. INTRODUCTION
19
Ertan YAKICI
logistics (Duran et al. [2], Yakıcı and Karasakal [3], Renkli and Duran [4]).
Naval operations like logistic support of warfare fleets, mine sweeping or
hunting operations is the basic motivation of this research. Type of logistic
support or operation, platform eligibility and other characteristics like
operation speed (or economic speed), desired support or operation periods
comprise several restrictions on the problem.
20
A Rich Min-Max Vehicle Routing Problem
2. PROBLEM DEFINITION
21
Ertan YAKICI
(1)
(2)
(3)
(4)
(5)
(6)
22
A Rich Min-Max Vehicle Routing Problem
(7)
(8)
(9)
, (10)
(11)
(12)
(13)
23
Ertan YAKICI
Note that still we can reduce the size of feasible space by demand
type cuts (14) which prevents the vehicles from visiting the demand points
that they cannot serve.
(14)
The heuristic approach is based on the idea of restriction of the MIP model.
Instead of solving the problem as a whole, we solve the problem in steps. At
each step, a group of vertices are added to the model until all of the vertices
are covered. The restricted problem is given a certain period of time for each
step. The traversed arcs of the best solution found is carried to the next step.
Each vehicle can break at most a limited number of arcs from its own route
where some (or all) of the carried arcs are allowed to be broken in order to
visit the vertices added. Another restriction is applied once in a while after a
certain number of steps are taken in between. This restriction fixes all of the
previously traversed arcs and service that are already processed in the
visited vertices. Therefore, the problem restarts again with the remaining
vertices. We call these steps “fixing steps”. After a fixing step, a vehicle
starts from the vertex it visits just before returning to the central vertex and
from the time it finishes its service in that last visited vertex. In order to
apply the restrictions, the following constraints are added to the model.
(15)
(16)
(17)
(18)
24
A Rich Min-Max Vehicle Routing Problem
25
Ertan YAKICI
In the selection of the arcs that are allowed to be broken, the number
of consecutive steps that an arc remain in the solutions is taken as the basic
parameter. If this parameter is set to “0”, all traversed arcs are allowed to be
broken in the next step. As the parameter increases, the number of arcs
allowed to be broken decreases. In all cases, the last arc of all vehicles
which is returning to central vertex is kept breakable. We accept that it is
hard to give a specific comment on the best combination for the remaining
parameters which are the number of added vertices at each step, the number
of steps between fixing steps and the time allowed for solving the restricted
problem. These parameters can be determined with respect to the problem
size and the time given to have a feasible solution.
(19)
(20)
26
A Rich Min-Max Vehicle Routing Problem
• Step 2. If L is not empty, add first nstep vertices (or add all vertices if there
are less than nstep) from L to model and update L, excluding the vertices
added to model, go to Step 3. Otherwise stop.
– Step 3.1. If this is not a fixing step and all vertices are served
without penalty, set the values to 1 for all traversed arcs (i,j) E
and k K where is equal to 1. Set to 1 w.r.t. given setting
in order to determine the arcs that can be broken, go to Step 2.
– Step 3.2. If this is a fixing step and all vertices are served without
penalty, go to Step 4.
– Step 3.3. If all vertices cannot be served, turn back to the last
fixing step and apply repair procedure. If solution cannot be repaired
go to previous fixing steps to apply repair procedure until the
penalized vertices are added to the model. Update L, go to Step 2.
• Step 4. Exclude all added vertices from model. For all excluded vertices;
keep the traversed arcs, served time windows and serving duration in the
vertices. Update the distances from central vertex to other vertices: For all
vehicles k K, set to where c indicates the last visited vertex of k.
Set the current vertices to “0”, and to for all vehicles
k K. Go to Step 2.
27
Ertan YAKICI
(21)
The binary parameter twniw level is set to positive value, if the vertex i
is served in time window w by any other vehicle that is not considered at the
current improvement step. While choosing the longest and shortest route,
the combination, giving the greatest number of vertices in longest route that
can be served by the vehicle of shortest route, is given priority. While the
set of longest routes consist of only the routes that have the maximum time
length, the set of shortest routes are populated starting from the route(s)
having shortest time length and adding one by one until the set satisfies that
the vehicles included are eligible to serve all vertices served by the set of
longest routes. To complete the improvement phase within reasonable time
period, each improvement iteration is given a certain period of time and
after allowing a certain number of steps (or time), when there is a
deterioration or no-improvement in the objective function, the algorithm is
stopped. The steps of the procedure applied for the improvement phase of
the heuristic approach is given below.
28
A Rich Min-Max Vehicle Routing Problem
• Step 2. Update the set of longest routes Rl and set of shortest routes Rs such
that Rl covers all rl longest routes having equal values, and Rs covers the
shortest rs routes of which vehicles can serve all of the vertices covered in
Rl. Go to Step 3.
• Step 3. If the stopping criterion does not hold, solve improvement model
with the combination of one route each from Rl and Rs, such that the number
of vertices in the longest route that can be served by the vehicle of the short
route be maximized (if there is a tie, choose the combination with shorter
route, if the tie is not broken, break randomly). Update the solution, and go
to Step 2. Otherwise stop.
4. EXPERIMENTS
29
Ertan YAKICI
30
A Rich Min-Max Vehicle Routing Problem
Observing Table 1 yields that we have rather poor lower bounds and
no solutions at all from commercial solver. When the problem is relaxed and
solved by the same solver, we can find better lower bounds which serves as
lower bounds to the original problem (MIP). Since the solver does not
perform well in finding solutions, we believe these lower bounds are still far
away from optimal solutions. Therefore, while we cannot determine the
quality of heuristic solution, we can claim that the heuristic is useful in this
case when there is no other algorithm that can find better solutions.
5. CONCLUSION
31
Ertan YAKICI
not have an idea about the solution quality, however we consider that the
proposed method is useful in the absence of a better solution.
REFERENCES
[1] Dantzig G.B., and Ramser J,H., (1959). The truck dispatching
problem, Management Science, vol. 6(1), pp. 80–91.
[2] Duran S., Gutierrez M.A., and Keskinocak P., (2011). Pre-
positioning of emergency items for care international, Interfaces, vol. 41(3),
pp. 223–237.
[3] Yakıcı E., and Karasakal, O., (2013). A minmax vehicle routing
problem with split delivery and heterogeneous demand, Optimization
Letters, vol. 7(7), pp. 1611–1625.
[4] Renkli Ç., and Duran S., (2015). Pre-positioning disaster response
facilities and relief items, Human and Ecological Risk Assessment: An
International Journal, vol. 21(5), pp. 1169–1185,.
[5] Drexl M., (2012). Synchronization in vehicle routing - a survey of
VRPs with multiple synchronization constraints, Transportation Science,
vol. 46(3), pp. 297–316.
[6] Bredström D., and Rönnqvist M., (2008). Combined vehicle routing
and scheduling with temporal precedence and synchronization constraints,
European Journal of Operational Research, vol. 191(1), pp. 19–31.
32